Move pixelTracksCPU module definition inside SwitchProducerCUDA

[makortel]

PR description:

While developing #36938 further (what is now in #37563), I came across this one case where a case-EDProducer of a SwitchProducer was duplicated between the python module (PixelTracks_cff) and SwitchProducerCUDA (pixelTracksSoA) instead of cloning the module. With my developments the duplication lead to failures in python. This PR clones the module when adding it in SwitchProducerCUDA. This PR moves the definition of what was pixelTracksCPU module inside the definition of pixelTracksSoA SwitchProducerCUDA.

PR validation:

The PixelTracks_cff file passes python with my developments.

[cmsbuild]

+code-checks

Logs: https://cmssdt.cern.ch/SDT/code-checks/cms-sw-PR-37562/29293

• This PR adds an extra 12KB to repository

[cmsbuild]

A new Pull Request was created by @makortel (Matti Kortelainen) for master.

It involves the following packages:

• RecoPixelVertexing/PixelTrackFitting (reconstruction)

@jpata, @cmsbuild, @clacaputo, @slava77 can you please review it and eventually sign? Thanks.
@felicepantaleo, @GiacomoSguazzoni, @JanFSchulte, @rovere, @VinInn, @mmusich, @mtosi, @dgulhan this is something you requested to watch as well.
@perrotta, @dpiparo, @qliphy you are the release manager for this.

cms-bot commands are listed here

[makortel]

@cmsbuild, please test

[makortel]

@cmsbuild, please abort

On further look I think this is not the correct fix. (edit: corrected in a later force-push)

[makortel]

The pixelTracksCPU appeared to be used only in this file, and this customization and the run3_common.toModify(pixelTracksSoA.cpu, on line 109/106 appeared to assume the pixelTracksCPU and pixelTracksSoA.cpu point to the same module object.

[makortel]

assign heterogeneous

[makortel]

enable gpu

[cmsbuild]

+code-checks

Logs: https://cmssdt.cern.ch/SDT/code-checks/cms-sw-PR-37562/29294

• This PR adds an extra 12KB to repository

[cmsbuild]

New categories assigned: heterogeneous

@fwyzard,@makortel you have been requested to review this Pull request/Issue and eventually sign? Thanks

[cmsbuild]

Pull request #37562 was updated. @makortel, @slava77, @clacaputo, @cmsbuild, @fwyzard, @jpata can you please check and sign again.

[jpata]

assign hlt

[cmsbuild]

New categories assigned: hlt

@missirol,@Martin-Grunewald you have been requested to review this Pull request/Issue and eventually sign? Thanks

[missirol]

+hlt

• the tests passed, and the fix seems uncontroversial
• not sure what I can add from HLT (isn't this part of an offline reco wf with 'heterogenous parts'? a priori unrelated to HLT?)

[fwyzard]

assign hlt

not sure why, as this is purely an offline reconstruction configuration ?

[jpata]

maybe I was confused, but I had the impression this runs at HLT / affects HLT. apologies if I was mistaken!

[jpata]

+reconstruction

• technical config reorganization, no changes

[fwyzard]

Ah... @jpata no, the HLT configuration is fully contained in the HLTrigger/Configuration/python files.
This should only be used by the xxx.502/xxx.506 offline workflows and their variations.

[fwyzard]

@makortel what did the previous version do ?
Did process.pixelTracksCPU and process.pixelTracksSoA@cpu identify the same module at the framework level, or two different modules ?

[makortel]

what did the previous version do ?
Did process.pixelTracksCPU and process.pixelTracksSoA@cpu identify the same module at the framework level, or two different modules ?

I think that they would lead to two modules being instantiated in the C++ side (except as things are process.pixelTracksCPU would not be instantiated in practice as long as it is not part of any Task/ConditionalTask/Sequence/Path/EndPath/Schedule).

In the python side, it would be some mixture of one and two modules

• any parameter change in either would be visible in other (as they are the same python object)
• process.pixelTracksSoA.cpu.label_() would return pixelTracksCPU
  • with this PR and before Make SwitchProducer work with ConditionalTask #37563 it would raise a RuntimeException with module has no label. Perhaps it wasn't inserted into the process?
    • after Make SwitchProducer work with ConditionalTask #37563 it will return pixelTracksSoA@cpu
• in config dump they would be visible as two separate modules (without the connection of being the same object)

[fwyzard]

Thanks for the summary and explanation !

[fwyzard]

+heterogeneous

[cmsbuild]

This pull request is fully signed and it will be integrated in one of the next master IBs (tests are also fine). This pull request will now be reviewed by the release team before it's merged. @perrotta, @dpiparo, @qliphy (and backports should be raised in the release meeting by the corresponding L2)

[qliphy]

+1